Multi-Contrast MRI Super-Resolution in Brain Tumors: Arbitrary-Scale Implicit Sampling and Unsupervised Fine-Tuning.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• Multi-contrast MRI provides comprehensive tissue characterization, valuable in clinical settings.
• Existing super-resolution (SR) methods for MRI face challenges due to equipment variability and lack of paired data.
• Clinical translation of multi-contrast MRI SR is limited by data distribution gaps and supervised training constraints.

Purpose of the Study:

• To develop a novel framework for unsupervised multi-contrast MRI super-resolution addressing clinical data challenges.
• To propose an implicit sampling and generation (ISG) network capable of arbitrary-scale SR.
• To introduce an unsupervised fine-tuning (FT) framework for adapting models to diverse clinical data.

Main Methods:

• Developed an Implicit Sampling and Generation (ISG) network for robust MRI super-resolution.
• Implemented an unsupervised fine-tuning (FT) framework as a test-time training technique.
• Validated the ISG+FT framework on clinical datasets including amide proton transfer weighted (APTw) and fluid-attenuated inversion recovery (FLAIR) images.

Main Results:

• The ISG+FT method achieved 4x super-resolution for APTw metabolic images in tumor patients.
• Demonstrated superior performance over state-of-the-art baselines in both quantitative and qualitative evaluations.
• Experimental results on diverse clinical datasets confirmed the method's effectiveness and robustness.

Conclusions:

• The proposed ISG+FT framework offers a promising solution for unsupervised multi-contrast MRI super-resolution in clinical practice.
• The method effectively overcomes limitations related to equipment discrepancies and data availability.
• This approach holds significant potential for improving diagnostic accuracy and clinical utility of MRI.